Top-down machine learning approach for high-throughput single-molecule analysis.

Single-molecule approaches provide enormous insight into the dynamics of biomolecules, but adequately sampling distributions of states and events often requires extensive sampling. Although emerging experimental techniques can generate such large datasets, existing analysis tools are not suitable to process the large volume of data obtained in high-throughput paradigms. Here, we present a new analysis platform (DISC) that accelerates unsupervised analysis of single-molecule trajectories. By merging model-free statistical learning with the Viterbi algorithm, DISC idealizes single-molecule trajectories up to three orders of magnitude faster with improved accuracy compared to other commonly used algorithms. Further, we demonstrate the utility of DISC algorithm to probe cooperativity between multiple binding events in the cyclic nucleotide binding domains of HCN pacemaker channel. Given the flexible and efficient nature of DISC, we anticipate it will be a powerful tool for unsupervised processing of high-throughput data across a range of single-molecule experiments. During a chemical or biological process, a molecule may transition through a series of states, many of which are rare or short-lived. Advances in technology have made it easier to detect these states by gathering large amounts of data on individual molecules. However, the increasing size of these datasets has put a strain on the algorithms and software used to identify different molecular states. Now, White et al. have developed a new algorithm called DISC which overcomes this technical limitation. Unlike most other algorithms, DISC requires minimal input from the user and uses a new method to group the data into categories that represent distinct molecular states. Although this new approach produces a similar end-result, it reaches this conclusion much faster than more commonly used algorithms. To test the effectiveness of the algorithm, White et al. studied how individual molecules of a chemical known as cAMP bind to parts of proteins called cyclic nucleotide binding domains (or CNDBs for short). A fluorescent tag was attached to single molecules of cAMP and data were collected on the behavior of each molecule. Previous evidence suggested that when four CNDBs join together to form a so-called tetramer complex, this affects the binding of cAMP. Using the DISC system, White et al. showed that individual cAMP molecules interact with all four domains in a similar way, suggesting that the binding of cAMP is not impacted by the formation of a tetramer complex. Analyzing this data took DISC less than 20 minutes compared to existing algorithms which took anywhere between four hours and two weeks to complete. The enhanced speed of the DISC algorithm could make it easier to analyze much larger datasets from other techniques in addition to fluorescence. This means that a greater number of states can be sampled, providing a deeper insight into the inner workings of biological and chemical processes.

Informations de copyright

© 2020, White et al.

Déclaration de conflit d'intérêts

DW, MG, RG No competing interests declared, BC Reviewing editor, eLife